News Sources, Journalists and Climate Change
Sylvan Zheng
- Basic Descriptives
- Ideological Sources
- Article Slant and Balance
- Journalist characteristics and article slant/balance
- More did specs
- Appendix
- Coverage/data quality
- Imputation: DIME Ideology is Correlated to Source Category
- Histogram of measured article ideology
- Histogram of predicted journalist ideology
- Distribution of source types by newspaper
- Bivariate analysis
- Citation Topics are Correlated to Source Types, Politician Parties, Ideology
- Non policy articles cite fewer ideological sources
- Alternate Specifications
- Investigating Drivers
- Elite Educated Slant Over Time
- NYT comment counts
- What articles get more comments?
- Journalist characteristics = journalist background is sort of significant I guess?
- More comments for more rightward leaning articles.. that’s basically it
- Not sure exactly how to interpret this…
- Interesting…
- no super intersting interaction with the post
# Load sources, articles, journos
datadir <- "data/masterdata2"
all.articles <- read_tsv(here(datadir, "articles.tsv"))
all.sources <- read_tsv(here(datadir, "sources.tsv"))
sources <- all.sources %>% filter(policy_label_gpt)
articles <- all.articles %>% filter(policy_label_gpt)
authors <- articles %>%
group_by(author_name, elite_undergrad_ivyplus, edu.undergrad, edu.has_postgrad, is_career, field.journo, age_est_2017, gender, race.nonwhite) %>%
summarize(n = n()) %>%
filter(!is.na(author_name))
articles.repna <- articles %>% mutate(
field.journo = replace_na(field.journo, FALSE),
elite_undergrad_ivyplus = replace_na(elite_undergrad_ivyplus, FALSE),
edu.has_postgrad = replace_na(edu.has_postgrad, FALSE),
)
sources.repna <- sources %>% mutate(
field.journo = replace_na(field.journo, FALSE),
elite_undergrad_ivyplus = replace_na(elite_undergrad_ivyplus, FALSE),
edu.has_postgrad = replace_na(edu.has_postgrad, FALSE),
)Basic Descriptives
Ns
nrow(articles)## [1] 13136nrow(sources)## [1] 77454nrow(authors)## [1] 1842Articles
Number Over Time By outlet
articles %>%
group_by(year, source) %>%
summarize(n = n()) %>%
ggplot(aes(x = year, y = n, color = source)) +
geom_line() +
theme_bw() +
scale_x_continuous(breaks = c(2012, 2014, 2016, 2018, 2020, 2022)) +
labs(
x = "Year",
y = "Number of Articles",
color = "Newspaper"
) +
theme(
panel.grid.minor = element_blank()
)
ggsave(here("paper/figures/time_trend_n_articles.png"), width = 6, height = 4)Sources
Number per Article
articles$n %>% hist()
summary(articles$n)## Min. 1st Qu. Median Mean 3rd Qu. Max.
## 1.000 4.000 6.000 5.896 8.000 31.000Number per Article over Time
sources %>%
group_by(year) %>%
summarize(n = n(), n_articles = length(unique(filename)), avg_sources = n / n_articles) %>%
ggplot(aes(x = year, y = avg_sources)) +
geom_line() +
scale_x_continuous(breaks = seq(2012, 2022, 2)) +
theme_bw()
sources %>%
group_by(year > 2016) %>%
summarize(n = n(), n_articles = length(unique(filename)), avg_sources = n / n_articles)## # A tibble: 2 × 4
## `year > 2016` n n_articles avg_sources
## <lgl> <int> <int> <dbl>
## 1 FALSE 17586 3336 5.27
## 2 TRUE 59868 9800 6.11Number per Outlet
sources %>%
group_by(source) %>%
summarize(n = n(), n_articles = length(unique(filename)), avg_sources = n / n_articles)## # A tibble: 6 × 4
## source n n_articles avg_sources
## <chr> <int> <int> <dbl>
## 1 Chicago Tribune 3052 578 5.28
## 2 Los Angeles Times 10173 1796 5.66
## 3 New York Times 28282 4671 6.05
## 4 USA Today 3397 684 4.97
## 5 Wall Street Journal 12081 2286 5.28
## 6 Washington Post 20469 3121 6.56sources %>%
group_by(source, year) %>%
summarize(n = n(), n_articles = length(unique(filename)), avg_sources = n / n_articles) %>%
ggplot(aes(x = year, y = avg_sources, color = source)) +
geom_line()
# The example articles
# filter(str_detect(filename, "1270322870.xml|1731405273.xml")) %>%Source Types (Table 1)
gen.cats <- table(sources$category) %>%
prop.table() %>%
as.data.frame() %>%
rename(Category = "Var1") %>%
mutate(
Freq = paste0(round(100 * Freq, digits = 1), "%")
)
sources$category2 <- case_when(
sources$env_category == "environmental" ~ paste(sources$category, "- Environmental"),
sources$env_category == "fossil fuel" ~ paste(sources$category, "- Fossil Fuel"),
sources$category == "Politician" & sources$pol_party == "Democrat" ~ "Politician - Democrat",
sources$category == "Politician" & sources$pol_party == "Republican" ~ "Politician - Republican",
sources$category == "Politician" & sources$category == "Politician" ~ "Politician - International",
sources$category == "Citizen" ~ "Other",
sources$category == "Advocacy" ~ "Advocacy - Other",
sources$category == "Business" ~ "Business - Other",
TRUE ~ sources$category
)
sub.cats <- table(sources$category2) %>%
prop.table() %>%
as.data.frame() %>%
rename(Category = "Var1") %>%
mutate(
Freq = paste0(round(100 * Freq, digits = 1), "%")
)
rbind(gen.cats, sub.cats) %>% distinct()## Category Freq
## 1 Academic 10.2%
## 2 Advocacy 21.5%
## 3 Bureaucrat 12.4%
## 4 Business 15%
## 5 International 15.2%
## 6 Media 5.6%
## 7 Other 4.1%
## 8 Politician 16%
## 9 Advocacy - Environmental 13.7%
## 10 Advocacy - Fossil Fuel 0.4%
## 11 Advocacy - Other 7.5%
## 12 Business - Environmental 2%
## 13 Business - Fossil Fuel 4.3%
## 14 Business - Other 8.6%
## 15 Politician - Democrat 9.7%
## 16 Politician - International 2.2%
## 17 Politician - Republican 4.1%Top Source DIME scores
dt <- sources %>%
filter(is.na(gov_category)) %>%
group_by(org_id) %>%
summarize(
n = n(), cfscore = mean(cfscore, na.rm = T), name = first(organization_name), category.slant = first(category.slant)
) %>%
arrange(desc(n)) %>%
filter(!is.na(cfscore)) %>%
head(50) %>%
arrange(n) %>%
mutate(rn = 51 - row_number(), n_size = n + 10)
dt %>% ggplot(aes(x = cfscore, size = sqrt(n), y = n, color = cfscore)) +
theme_bw() +
# scale_y_continuous(limits = c(-1, 0.8)) +
scale_y_log10() +
# scale_y_continuous(trans='reverse') +
scale_x_continuous(limits = c(-1.5, 1.2)) +
scale_color_gradient(low = "blue", high = "red") +
geom_vline(xintercept = 0, lty = "dashed") +
theme(panel.grid = element_blank(), axis.text.y = element_blank()) +
labs(x = "DIME Ideology", y = "") +
geom_text_repel(
data = subset(dt, category.slant != "other" & !(name %in% c("The Associated Press", "Associated Press", "California"))),
aes(x = cfscore, label = name),
direction = "y",
min.segment.length = 999
) +
guides(size = "none", color = "none", fill = "none")
ggsave(here("paper/figures/dime-scatter-100.png"), width = 10, height = 6)# scratchTop DIME over time
dt <- sources %>%
filter(is.na(gov_category)) %>%
group_by(org_id) %>%
summarize(
n = n(), cfscore = mean(cfscore, na.rm = T), name = first(organization_name), category.slant = first(category.slant)
) %>%
arrange(desc(n)) %>%
filter(!is.na(cfscore)) %>%
head(7)
dt## # A tibble: 7 × 5
## org_id n cfscore name category.slant
## <dbl> <int> <dbl> <chr> <chr>
## 1 3604 1288 0.265 United States of America Democrat
## 2 9187 786 0.287 United States Government Republican
## 3 3487 656 -1.04 Natural Resources Defense Council (NRDC) Environmental
## 4 5505 430 -1.10 Environmental Defense Fund Environmental
## 5 1072 389 1.01 International Energy Agency Advocacy
## 6 10671 384 0.193 American Petroleum Institute Fossil Fuel
## 7 1620 345 -0.960 Sierra Club Environmentalsources %>% filter(org_id %in% dt$org_id) %>% group_by(year, organization_name, org_id, cfscore) %>% summarize(n=n()) %>% ggplot(aes(
x=year, y=n, color=cfscore < 0, lty=organization_name
)) +
geom_smooth(se=F)
Politicians
pres <- sources %>%
mutate(
pres = case_when(
str_detect(person_name, "Biden") ~ "Biden",
str_detect(person_name, "Obama") ~ "Obama",
str_detect(person_name, "Trump") ~ "Trump",
category.slant %in% c("Democrat", "Republican") ~ category.slant,
TRUE ~ NA
)
)
npy <- sources %>%
group_by(year) %>%
summarize(npy = n())
pres %>%
group_by(year, pres) %>%
filter(!is.na(pres)) %>%
summarize(n = n()) %>%
ggplot(aes(x = year, y = n, color = pres)) +
geom_line() +
theme_bw()
# Normalize by # of sources per year...
pres %>%
group_by(year, pres) %>%
filter(!is.na(pres)) %>%
summarize(n = n()) %>%
left_join(npy) %>%
ggplot(aes(x = year, y = n / npy, color = pres, shape = pres)) +
geom_line() +
geom_point() +
scale_x_continuous(breaks = seq(2012, 2022, 4)) +
theme_bw()
Journalists
Demographics
ggarrange(plotlist = list(
authors %>% ggplot(aes(x = age_est_2017)) +
geom_histogram(),
authors %>% ggplot(aes(x = elite_undergrad_ivyplus)) +
geom_bar(stat = "count"),
authors %>% ggplot(aes(x = edu.has_postgrad)) +
geom_bar(stat = "count"),
authors %>% ggplot(aes(x = field.journo)) +
geom_bar(stat = "count"),
authors %>% ggplot(aes(x = gender)) +
geom_bar(stat = "count"),
authors %>% ggplot(aes(x = race.nonwhite)) +
geom_bar(stat = "count")
))
Demographics over time
Unit of analysis: unique journalists per year
journo.years <- articles %>%
group_by(author_name, year, elite_undergrad_ivyplus, edu.undergrad, edu.has_postgrad, is_career, field.journo, age_est, gender, race.nonwhite) %>%
summarize(n = n()) %>%
filter(!is.na(author_name))
age.plt <- journo.years %>%
ggplot(aes(x = year, y = as.numeric(age_est > 35))) +
geom_smooth(method = "loess") +
labs(x = "Year", y = "Age > 35")
ivy.plt <- journo.years %>%
ggplot(aes(x = year, y = as.numeric(elite_undergrad_ivyplus))) +
geom_smooth(method = "loess") +
labs(y = "Ivy League", x = "Year")
fj.plt <- journo.years %>%
ggplot(aes(x = year, y = as.numeric(field.journo))) +
labs(y = "Journalism Degree", x = "Year") +
geom_smooth(method = "loess")
pg.plt <- journo.years %>%
ggplot(aes(x = year, y = as.numeric(edu.has_postgrad))) +
labs(y = "Postgraduate Degree", x = "Year") +
geom_smooth(method = "loess")
gender.plt <- journo.years %>%
ggplot(aes(x = year, y = as.numeric(gender == "female"))) +
labs(y = "Female", x = "Year") +
geom_smooth(method = "loess")
race.plt <- journo.years %>%
ggplot(aes(x = year, y = as.numeric(race.nonwhite))) +
labs(y = "Nonwhite", x = "Year") +
geom_smooth(method = "loess")
plts <- list(age.plt, ivy.plt, fj.plt, gender.plt, pg.plt, race.plt) %>%
lapply(function(p) {
p +
theme_bw() +
theme(panel.grid = element_blank()) +
scale_x_continuous(breaks = seq(2012, 2022, 2))
})
ggarrange(plotlist = plts)
ggsave(here("paper/figures/journo-agg-2.png"), width = 10, height = 6)Ideological Sources
articles %>%
pivot_longer(cols = c("n_right_dime_other", "n_left_dime_other", "n_rep", "n_dem", "n_env", "n_ff", "n_center_total")) %>%
ggplot(aes(x = date, y = value, color = name)) +
geom_smooth(se = F) +
theme_bw()
sources %>%
group_by(year, category.slant) %>%
summarize(n = n()) %>%
ggplot(aes(x = year, y = n, color = category.slant, lty = category.slant)) +
geom_smooth(se = F) +
theme_bw()
sources_per_year <- sources %>%
group_by(year) %>%
summarize(year.n = n())
sources %>%
group_by(year, category.slant) %>%
summarize(n = n()) %>%
left_join(sources_per_year) %>%
ggplot(aes(x = year, y = n / year.n, color = category.slant, lty = category.slant)) +
geom_line() +
theme_bw()
Article Slant and Balance
Slant/Balance Histogram
ggplot(articles, aes(x = ideo.mean.i, fill = bal.at_least_one_both_sides)) +
geom_histogram() +
theme_bw()
ggplot(articles, aes(x = bal.diff_lr_normalized_nonabs, fill = bal.at_least_one_both_sides)) +
geom_histogram() +
theme_bw()
Slant and Balance Over Time
# With loess
slant_overview_plt <- articles %>%
pivot_longer(
cols = c("ideo.mean.i", "ideo.mean.pols", "ideo.mean.i.orgs"),
) %>%
mutate(
name = dplyr::recode(name, ideo.mean.i = "All", "ideo.mean.pols" = "Politicians", "ideo.mean.i.orgs" = "Organizations"),
size = case_when(name == "All" ~ 1.5, TRUE ~ 1),
lty = case_when(name == "All" ~ 1, TRUE ~ 2)
) %>%
ggplot(aes(x = date, y = value, color = name, lty = as.factor(lty))) +
geom_smooth(se = F, method = "loess") +
theme_bw() +
theme(
panel.grid.minor = element_blank(),
panel.grid.major = element_blank(),
) +
guides(
linetype = "none",
size = "none"
) +
labs(
y = "Mean article slant",
x = "Year",
color = "Slant Type"
)balance_plt <- articles %>%
pivot_longer(cols = c("balance_all", "balance_ff_env", "balance_rep_dem")) %>%
mutate(
name = dplyr::recode(name, balance_all = "All", balance_ff_env = "Organizations", balance_rep_dem = "Politicians"),
size = case_when(name == "All" ~ 1.5, TRUE ~ 1),
lty = case_when(name == "All" ~ 1, TRUE ~ 2)
) %>%
ggplot(aes(x = date, y = as.numeric(value), color = name, lty = as.factor(lty))) +
geom_smooth(se = F, method = "loess") +
theme_bw() +
theme(
panel.grid.minor = element_blank(),
panel.grid.major = element_blank(),
) +
guides(
linetype = "none",
size = "none"
) +
labs(
y = "Proportion of balanced articles",
x = "Year",
color = "Slant Type"
)
ggarrange(plotlist = list(slant_overview_plt, balance_plt), common.legend = T, legend = "bottom")
ggsave(here("paper/figures/slant-balance-overview.png"), width = 8, height = 5)Article Balance (Ratio)
articles %>% ggplot(aes(x = date, y = as.numeric(bal.diff_lr_normalized > 0))) +
geom_smooth() +
theme_bw()
Slant SD
articles$ideo.sd.i %>% hist
ggplot(articles, aes(x=date, y=ideo.sd.i)) + geom_smooth()
# Really interesting shape, although the magnitude is maybe not as large?
# Hard to interpret move from 0.65SD peak to 0.55 dip
ggarrange(plotlist=list(
ggplot(articles, aes(x=date, y=ideo.sd.orgs)) + geom_smooth(),
ggplot(articles, aes(x=date, y=ideo.sd.pols)) + geom_smooth()
))
# Mostly driven by orgs...
ggplot(articles, aes(x=date, y=ideo.sd.i, color=source)) + geom_smooth(se=F)
# Wow that's kinda interesting for sureSlant over Time by Outlet
articles %>% ggplot(aes(x = date, y = ideo.mean.i, color = source)) +
geom_smooth(se = F, method = "loess") +
labs(title = "Article slant by average organization DIME score")
Business vs Environmental Groups
articles %>% ggplot(aes(x = date, y = ideo.mean.i.orgs, color = source)) +
geom_smooth(se = F, method = "loess")
Democrat and Republican Politicians
articles %>% ggplot(aes(x = date, y = ideo.mean.pols, color = source)) +
geom_smooth(se = F, method = "loess")
Journalist characteristics and article slant/balance
make_fmla <- function(y, covariates, festring = "| year + source") {
covariate_str <- paste0(covariates, collapse = "+")
as.formula(paste(y, "~", covariate_str, festring))
}
make_models <- function(df, yvars, covariates, festring = "| year + source", cluster = "author_name") {
lapply(yvars, \(y) {
feols(make_fmla(y, covariates, festring), data = df, cluster = cluster)
})
}
non_edu_covars <- c(
"(age_est_2017 > 30)",
"gender",
"race.nonwhite"
)
j_covars <- c(
"elite_undergrad_ivyplus",
"edu.has_postgrad",
"field.journo",
non_edu_covars
)
models <- c(
make_models(articles, c("n", "n_unique_source_category"), j_covars),
make_models(articles, c("balance_all", "ideo.mean.i", "ideo.sd.i"), c(j_covars, "n"))
)
etable(models)## model 1 model 2
## Dependent Var.: n n_unique_source_category
##
## elite_undergrad_ivyplusTRUE 0.3623* (0.1578) 0.1872* (0.0771)
## edu.has_postgradTRUE -0.1035 (0.1365) 0.0128 (0.0684)
## field.journoTRUE -0.1710 (0.1317) -0.0382 (0.0658)
## age_est_2017>30TRUE -0.1442 (0.1783) 0.0804 (0.1049)
## gendermale 0.0880 (0.1281) -0.0423 (0.0706)
## race.nonwhiteTRUE 0.1511 (0.1690) -0.0275 (0.0765)
## n
## Fixed-Effects: ---------------- ------------------------
## year Yes Yes
## source Yes Yes
## ___________________________ ________________ ________________________
## S.E.: Clustered by: author_name by: author_name
## Observations 9,349 9,349
## R2 0.05600 0.04279
## Within R2 0.00621 0.00653
##
## model 3 model 4
## Dependent Var.: balance_all ideo.mean.i
##
## elite_undergrad_ivyplusTRUE 0.1335*** (0.0262) 0.0888*** (0.0193)
## edu.has_postgradTRUE -0.0285 (0.0255) -0.0223 (0.0188)
## field.journoTRUE 0.1060*** (0.0276) 0.0627** (0.0221)
## age_est_2017>30TRUE 0.0345 (0.0249) 0.0518** (0.0180)
## gendermale -0.0153 (0.0226) -0.0122 (0.0174)
## race.nonwhiteTRUE -0.0773. (0.0421) -0.0716* (0.0303)
## n 0.0500*** (0.0028) 0.0076*** (0.0021)
## Fixed-Effects: ------------------ ------------------
## year Yes Yes
## source Yes Yes
## ___________________________ __________________ __________________
## S.E.: Clustered by: author_name by: author_name
## Observations 8,449 8,636
## R2 0.11423 0.07208
## Within R2 0.10020 0.01644
##
## model 5
## Dependent Var.: ideo.sd.i
##
## elite_undergrad_ivyplusTRUE 0.0764*** (0.0212)
## edu.has_postgradTRUE -0.0128 (0.0197)
## field.journoTRUE 0.0505** (0.0176)
## age_est_2017>30TRUE 0.0258 (0.0182)
## gendermale -0.0477** (0.0174)
## race.nonwhiteTRUE -0.0643*** (0.0187)
## n 0.0054* (0.0023)
## Fixed-Effects: -------------------
## year Yes
## source Yes
## ___________________________ ___________________
## S.E.: Clustered by: author_name
## Observations 7,231
## R2 0.04577
## Within R2 0.02405
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1etable(c(
make_models(articles.repna, c("balance_all", "ideo.mean.i", "ideo.sd.i"), c("elite_undergrad_ivyplus*gender", "n"))
))## model 1 model 2
## Dependent Var.: balance_all ideo.mean.i
##
## elite_undergrad_ivyplusTRUE 0.1743*** (0.0380) 0.1064*** (0.0311)
## gendermale 0.0471 (0.0287) 0.0274 (0.0234)
## n 0.0510*** (0.0025) 0.0081*** (0.0021)
## elite_undergrad_ivyplusTRUE x gendermale -0.1454** (0.0463) -0.0652. (0.0378)
## Fixed-Effects: ------------------ ------------------
## year Yes Yes
## source Yes Yes
## ________________________________________ __________________ __________________
## S.E.: Clustered by: author_name by: author_name
## Observations 10,713 11,024
## R2 0.10376 0.06133
## Within R2 0.09276 0.00788
##
## model 3
## Dependent Var.: ideo.sd.i
##
## elite_undergrad_ivyplusTRUE 0.1233*** (0.0286)
## gendermale 0.0052 (0.0167)
## n 0.0062** (0.0019)
## elite_undergrad_ivyplusTRUE x gendermale -0.1039** (0.0345)
## Fixed-Effects: ------------------
## year Yes
## source Yes
## ________________________________________ __________________
## S.E.: Clustered by: author_name
## Observations 9,152
## R2 0.03601
## Within R2 0.01858
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1# Basically same results with sources as the unit of analysis
c(
make_models(sources, c("cfscore"), c(j_covars)),
make_models(sources.repna, c("cfscore"), c(j_covars))
) %>% etable()## model 1 model 2
## Dependent Var.: cfscore cfscore
##
## elite_undergrad_ivyplusTRUE 0.0948*** (0.0256) 0.1010*** (0.0244)
## edu.has_postgradTRUE -0.0307 (0.0239) -0.0199 (0.0230)
## field.journoTRUE 0.0209 (0.0316) 0.0352 (0.0282)
## age_est_2017>30TRUE 0.0701*** (0.0206) 0.0699*** (0.0196)
## gendermale -0.0113 (0.0217) -0.0171 (0.0208)
## race.nonwhiteTRUE -0.0396 (0.0414) -0.0257 (0.0394)
## Fixed-Effects: ------------------ ------------------
## year Yes Yes
## source Yes Yes
## ___________________________ __________________ __________________
## S.E.: Clustered by: author_name by: author_name
## Observations 18,459 21,225
## R2 0.03510 0.03238
## Within R2 0.00452 0.00450
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1Alternate Specification: No journalism field
j_covars2 <- c(
"elite_undergrad_ivyplus",
"edu.has_postgrad",
non_edu_covars
)
models.alt.nj <- c(
make_models(articles, c("n", "n_unique_source_category"), j_covars2),
make_models(articles, c("balance_all", "ideo.mean.i"), c(j_covars2, "n"))
)
etable(models.alt.nj)## model 1 model 2
## Dependent Var.: n n_unique_source_category
##
## elite_undergrad_ivyplusTRUE 0.2944* (0.1437) 0.1481* (0.0752)
## edu.has_postgradTRUE -0.0584 (0.1247) 0.0230 (0.0664)
## age_est_2017>30TRUE -0.0852 (0.1756) 0.0791 (0.1009)
## gendermale -0.0309 (0.1245) -0.0731 (0.0659)
## race.nonwhiteTRUE 0.0873 (0.1662) -0.0194 (0.0746)
## n
## Fixed-Effects: ---------------- ------------------------
## year Yes Yes
## source Yes Yes
## ___________________________ ________________ ________________________
## S.E.: Clustered by: author_name by: author_name
## Observations 10,612 10,612
## R2 0.05368 0.04222
## Within R2 0.00286 0.00434
##
## model 3 model 4
## Dependent Var.: balance_all ideo.mean.i
##
## elite_undergrad_ivyplusTRUE 0.0787** (0.0272) 0.0595** (0.0194)
## edu.has_postgradTRUE -0.0378 (0.0270) -0.0344. (0.0193)
## age_est_2017>30TRUE 0.0274 (0.0264) 0.0470* (0.0186)
## gendermale -0.0215 (0.0255) -0.0049 (0.0191)
## race.nonwhiteTRUE -0.0703. (0.0411) -0.0748** (0.0285)
## n 0.0505*** (0.0026) 0.0069*** (0.0021)
## Fixed-Effects: ------------------ ------------------
## year Yes Yes
## source Yes Yes
## ___________________________ __________________ __________________
## S.E.: Clustered by: author_name by: author_name
## Observations 9,550 9,771
## R2 0.10337 0.06549
## Within R2 0.09119 0.01200
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1Alternate Specification: NA edu fields set to default
articles2 <- articles %>% mutate(
field.journo = replace_na(field.journo, FALSE),
elite_undergrad_ivyplus = replace_na(elite_undergrad_ivyplus, FALSE),
edu.has_postgrad = replace_na(edu.has_postgrad, FALSE),
)
models.alt.repna <- c(
make_models(articles2, c("n", "n_unique_source_category"), j_covars),
make_models(articles2, c("balance_all", "ideo.mean.i"), c(j_covars, "n"))
)
models.alt.repna.nofe <- c(
make_models(articles2, c("n", "n_unique_source_category"), j_covars, festring = "| year"),
make_models(articles2, c("balance_all", "ideo.mean.i"), c(j_covars, "n"), festring = "| year")
)
etable(models.alt.repna)## model 1 model 2
## Dependent Var.: n n_unique_source_category
##
## elite_undergrad_ivyplusTRUE 0.2891. (0.1488) 0.1619* (0.0708)
## edu.has_postgradTRUE -0.0737 (0.1255) 0.0255 (0.0615)
## field.journoTRUE -0.1119 (0.1211) -0.0018 (0.0633)
## age_est_2017>30TRUE -0.0114 (0.1673) 0.0959 (0.0927)
## gendermale -0.0273 (0.1206) -0.0733 (0.0636)
## race.nonwhiteTRUE 0.0902 (0.1598) -0.0173 (0.0714)
## n
## Fixed-Effects: ---------------- ------------------------
## year Yes Yes
## source Yes Yes
## ___________________________ ________________ ________________________
## S.E.: Clustered by: author_name by: author_name
## Observations 11,139 11,139
## R2 0.05525 0.04316
## Within R2 0.00345 0.00520
##
## model 3 model 4
## Dependent Var.: balance_all ideo.mean.i
##
## elite_undergrad_ivyplusTRUE 0.1173*** (0.0263) 0.0810*** (0.0193)
## edu.has_postgradTRUE -0.0142 (0.0255) -0.0148 (0.0184)
## field.journoTRUE 0.1082*** (0.0250) 0.0614** (0.0208)
## age_est_2017>30TRUE 0.0276 (0.0237) 0.0507** (0.0167)
## gendermale -0.0267 (0.0234) -0.0126 (0.0176)
## race.nonwhiteTRUE -0.0563 (0.0390) -0.0670* (0.0275)
## n 0.0513*** (0.0026) 0.0074*** (0.0021)
## Fixed-Effects: ------------------ ------------------
## year Yes Yes
## source Yes Yes
## ___________________________ __________________ __________________
## S.E.: Clustered by: author_name by: author_name
## Observations 9,977 10,243
## R2 0.11047 0.06560
## Within R2 0.09869 0.01402
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1Coefficient Graph for all specifications
all.models <- c(models, models.alt.nj, models.alt.repna, models.alt.repna.nofe)
plot_models <- function(models, title) {
mtypes <- c("Standard", "No Education Field", "Replace NAs with Defaults", "No Newspaper Fixed Effect")
ggcoefplot(as.list(models)) +
coord_flip() +
scale_color_discrete(labels = mtypes) +
scale_shape_discrete(labels = mtypes) +
labs(title = title) +
theme(
panel.grid.major = element_blank(),
panel.grid.minor = element_blank()
) +
guides(fill = "none")
}
ggarrange(
plotlist = list(
plot_models(all.models[seq(1, length(all.models), 4)], title = "Number of Sources"),
plot_models(all.models[seq(2, length(all.models), 4)], title = "Number of Unique Source Types"),
plot_models(all.models[seq(3, length(all.models), 4)], title = "Balance"),
plot_models(all.models[seq(4, length(all.models), 4)], title = "Slant")
),
common.legend = T,
legend = "bottom"
)
ggsave(here("paper/figures/coefplot-robust.png"), width = 12, height = 15)etable(models.alt.repna, tex = TRUE, digits = 2, fitstat = c("n"))\begingroup
\centering
\begin{tabular}{lcccc}
\tabularnewline \midrule \midrule
Dependent Variables: & n & n\_unique\_source\_category & balance\_all & ideo.mean.i\\
Model: & (1) & (2) & (3) & (4)\\
\midrule
\emph{Variables}\\
elite\_undergrad\_ivyplusTRUE & 0.29$^{*}$ & 0.16$^{**}$ & 0.12$^{***}$ & 0.08$^{***}$\\
& (0.15) & (0.07) & (0.03) & (0.02)\\
edu.has\_postgradTRUE & -0.07 & 0.03 & -0.01 & -0.01\\
& (0.13) & (0.06) & (0.03) & (0.02)\\
field.journoTRUE & -0.11 & -0.002 & 0.11$^{***}$ & 0.06$^{***}$\\
& (0.12) & (0.06) & (0.03) & (0.02)\\
age\_est\_2017>30TRUE & -0.01 & 0.10 & 0.03 & 0.05$^{***}$\\
& (0.17) & (0.09) & (0.02) & (0.02)\\
gendermale & -0.03 & -0.07 & -0.03 & -0.01\\
& (0.12) & (0.06) & (0.02) & (0.02)\\
race.nonwhiteTRUE & 0.09 & -0.02 & -0.06 & -0.07$^{**}$\\
& (0.16) & (0.07) & (0.04) & (0.03)\\
n & & & 0.05$^{***}$ & 0.007$^{***}$\\
& & & (0.003) & (0.002)\\
\midrule
\emph{Fixed-effects}\\
year & Yes & Yes & Yes & Yes\\
source & Yes & Yes & Yes & Yes\\
\midrule
\emph{Fit statistics}\\
Observations & 11,139 & 11,139 & 9,977 & 10,243\\
\midrule \midrule
\multicolumn{5}{l}{\emph{Clustered (author\_name) standard-errors in parentheses}}\\
\multicolumn{5}{l}{\emph{Signif. Codes: ***: 0.01, **: 0.05, *: 0.1}}\\
\end{tabular}
\par\endgroupCoefficient plot for best specification
ggcoefplot(as.list(c(models[1], models[2]))) +
# scale_x_discrete(limits = rev, labels = rev(c("Elite Undergrad", "Postgrad", "Journalism Degree", "Age > 30", "Male", "Non White", "Number of Sources"))) +
# scale_color_discrete(labels=c('Article Slant', 'Article Balance')) +
# scale_shape_discrete(labels=c('Article Slant', 'Article Balance')) +
coord_flip() +
labs(title = "") +
theme(
panel.grid.major = element_blank(),
panel.grid.minor = element_blank()
) +
guides(fill = "none")
ggcoefplot(as.list(c(models[3], models[4]))) +
# scale_x_discrete(limits = rev, labels = rev(c("Elite Undergrad", "Postgrad", "Journalism Degree", "Age > 30", "Male", "Non White", "Number of Sources"))) +
# scale_color_discrete(labels=c('Article Slant', 'Article Balance')) +
# scale_shape_discrete(labels=c('Article Slant', 'Article Balance')) +
coord_flip() +
labs(title = "") +
theme(
panel.grid.major = element_blank(),
panel.grid.minor = element_blank()
) +
guides(fill = "none")
ggsave(here("paper/figures/main-coefplot.png"), width = 6, height = 7)Robustness check to age threshold
vars <- c(
"gender",
"race.nonwhite",
"elite_undergrad_ivyplus",
"edu.has_postgrad",
"field.journo"
)
est_with_age_thresh <- function(thresh) {
var <- paste0("(age_est_2017 > ", thresh, ")")
models <- make_models(articles, c("ideo.mean.i", "balance_all"), c(vars, var, "n"))
ideo.age <- models[[1]]$coeftable[6, ]
balance.age <- models[[2]]$coeftable[6, ]
res <- as.data.frame(rbind(ideo.age, balance.age))
res$yvar <- c("ideo.mean.i", "balance_all")
res$thresh <- thresh
res
}
res <- bind_rows(lapply(seq(20, 50), est_with_age_thresh))
res %>% ggplot(aes(x = thresh, y = Estimate, color = yvar, ymin = Estimate - 1.96 * `Std. Error`, ymax = Estimate + 1.96 * `Std. Error`)) +
geom_hline(yintercept = 0, lty = 2) +
geom_point() +
geom_errorbar(position = position_dodge()) +
theme_bw()
ggsave(here("paper/figures/age_robustness.png"), width = 8, height = 6)Journalist Characteristics and Balance Ratio
ts <- seq(0, 0.95, 0.05)
ms <- sapply(ts, \(t) paste0("bal.diff_lr_normalized>", t))
tmodels <- make_models(articles2, ms, c(j_covars))
coefs <- lapply(ts, \(t) {
df <- as.data.frame(tmodels[[t * 10 + 1]]$coeftable)
df$t <- t
df$x <- rownames(df)
df
}) %>% bind_rows()
ggplot(coefs, aes(x = t, y = `Estimate`, ymin = `Estimate` - 1.96 * `Std. Error`, ymax = `Estimate` + 1.96 * `Std. Error`, color = `Pr(>|t|)` < 0.05)) +
geom_point() +
geom_errorbar(width = 0) +
geom_hline(yintercept = 0) +
theme_bw() +
facet_wrap(~x) +
labs()
# Journalism vaguely associated with higher diff
# Older less associated
# Ivyleaguers are more likely to be slightly biased but not super biased
# Same for ninwhites
# No gender effect
# I guess the other thing we should look at is whether its a left or right imbalance...
ts <- seq(-0.95, 0.95, 0.1)
ms <- sapply(ts, \(t) paste0("bal.diff_lr_normalized_nonabs", ifelse(t < 0, " < ", " > "), t))
feols(bal.diff_lr_normalized_nonabs < -0.5 ~ elite_undergrad_ivyplus, data = articles2)## OLS estimation, Dep. Var.: bal.diff_lr_normalized_nonabs < -0.5
## Observations: 13,136
## Standard-errors: IID
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) 0.024322 0.001560 15.59480 < 2.2e-16 ***
## elite_undergrad_ivyplusTRUE -0.006350 0.002767 -2.29462 0.021771 *
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
## RMSE: 0.147644 Adj. R2: 3.246e-4tmodels <- make_models(articles2, ms, c(j_covars))
coefs <- lapply(seq(1, length(ts)), \(t) {
df <- as.data.frame(tmodels[[t]]$coeftable)
df$t <- ts[[t]]
df$x <- rownames(df)
df
}) %>% bind_rows()
ggplot(coefs, aes(x = t, y = `Estimate`, ymin = `Estimate` - 1.96 * `Std. Error`, ymax = `Estimate` + 1.96 * `Std. Error`, color = `Pr(>|t|)` < 0.05)) +
geom_vline(xintercept = 0, lty = 2) +
geom_point() +
geom_errorbar(width = 0) +
geom_hline(yintercept = 0) +
theme_bw() +
scale_x_reverse() +
scale_color_manual(values = c("#888888", "cyan3")) +
facet_wrap(~x) +
labs()
# So this tells us that... Ivy league Js and older Js are more likely to write
# articles that include more right leaning sources.. but not a LOT more just a bit more
# This comes from the estimates closer to 0 as the threshold moves down
# Older J's are also less likely to write articles with more left leaning sources
# Nonwhites are almost more likely to write aticles with a couple more left leaning sources (not statsig though)
# Okay and then another possibility we can look at is basically
# who is more likely to write an article that is actually unbalanced but looks balanced
make_models(articles2, c(
"bal.diff_lr_normalized > 0.25 & bal.at_least_one_both_sides",
"abs(ideo.mean.i) > 0.25 & bal.at_least_one_both_sides"
), c(j_covars))## [[1]]
## OLS estimation
## Dep. Var.: bal.diff_lr_normalized > 0.25 & bal.at_least_one_both_sides
## Observations: 11,139
## Fixed-effects: year: 11, source: 6
## Standard-errors: Clustered (author_name)
## Estimate Std. Error t value Pr(>|t|)
## elite_undergrad_ivyplusTRUE 0.041877 0.015923 2.629916 0.0086471 **
## edu.has_postgradTRUE -0.010532 0.014763 -0.713420 0.4757211
## field.journoTRUE 0.052485 0.017470 3.004324 0.0027155 **
## age_est_2017 > 30TRUE -0.000146 0.017357 -0.008400 0.9932991
## gendermale -0.008573 0.014303 -0.599356 0.5490457
## race.nonwhiteTRUE -0.014844 0.019459 -0.762844 0.4457026
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
## RMSE: 0.350077 Adj. R2: 0.010026
## Within R2: 0.005839
##
## [[2]]
## OLS estimation
## Dep. Var.: abs(ideo.mean.i) > 0.25 & bal.at_least_one_both_sides
## Observations: 10,422
## Fixed-effects: year: 11, source: 6
## Standard-errors: Clustered (author_name)
## Estimate Std. Error t value Pr(>|t|)
## elite_undergrad_ivyplusTRUE 0.067097 0.016263 4.125648 3.9561e-05 ***
## edu.has_postgradTRUE -0.004148 0.015356 -0.270097 7.8713e-01
## field.journoTRUE 0.060465 0.016996 3.557649 3.8902e-04 ***
## age_est_2017 > 30TRUE 0.011341 0.016768 0.676339 4.9896e-01
## gendermale -0.010632 0.014772 -0.719717 4.7184e-01
## race.nonwhiteTRUE -0.033952 0.021383 -1.587821 1.1259e-01
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
## RMSE: 0.40864 Adj. R2: 0.012834
## Within R2: 0.008105# Conditional on writing a biased article, elite and journos are more likely
# to include an opposing view. this could mean they are more sophisitcated essentially?
make_models(articles2 %>% filter(bal.diff_lr_normalized > 0), c(
"bal.at_least_one_both_sides"
), c(j_covars))## [[1]]
## OLS estimation, Dep. Var.: bal.at_least_one_both_sides
## Observations: 8,807
## Fixed-effects: year: 11, source: 6
## Standard-errors: Clustered (author_name)
## Estimate Std. Error t value Pr(>|t|)
## elite_undergrad_ivyplusTRUE 0.110242 0.028819 3.825271 0.00013808 ***
## edu.has_postgradTRUE -0.014989 0.027862 -0.537957 0.59071706
## field.journoTRUE 0.090676 0.026395 3.435335 0.00061423 ***
## age_est_2017 > 30TRUE 0.043879 0.028358 1.547326 0.12207662
## gendermale -0.029213 0.027101 -1.077940 0.28130044
## race.nonwhiteTRUE -0.052514 0.038582 -1.361096 0.17376623
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
## RMSE: 0.47676 Adj. R2: 0.028732
## Within R2: 0.017153make_models(articles2 %>% filter(bal.diff_lr_normalized == 0), c(
"!is.na(bal.at_least_one_both_sides)"
), c(j_covars))## [[1]]
## OLS estimation, Dep. Var.: !is.na(bal.at_least_one_both_sides)
## Observations: 2,332
## Fixed-effects: year: 11, source: 6
## Standard-errors: Clustered (author_name)
## Estimate Std. Error t value Pr(>|t|)
## elite_undergrad_ivyplusTRUE 0.282633 0.050936 5.548803 4.2194e-08 ***
## edu.has_postgradTRUE 0.015394 0.043008 0.357921 7.2052e-01
## field.journoTRUE 0.181502 0.043759 4.147724 3.8126e-05 ***
## age_est_2017 > 30TRUE 0.069850 0.044404 1.573059 1.1620e-01
## gendermale -0.041185 0.040233 -1.023663 3.0638e-01
## race.nonwhiteTRUE 0.011522 0.047440 0.242885 8.0817e-01
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
## RMSE: 0.47712 Adj. R2: 0.081138
## Within R2: 0.067293ggplot(articles2, aes(x = bal.diff_lr_normalized_nonabs, fill = bal.at_least_one_both_sides)) +
geom_histogram() +
facet_wrap(~elite_undergrad_ivyplus) +
scale_x_reverse()
# Yeah so the ivy league ppl basically are more likely to include both sides
# Reverse isn't true - conditioning on including both sides doesn't increase likelihood of slant
make_models(articles2 %>% filter(bal.at_least_one_both_sides), c(
"bal.diff_lr_normalized > 0.25"
), c(j_covars))## [[1]]
## OLS estimation, Dep. Var.: bal.diff_lr_normalized > 0.25
## Observations: 4,480
## Fixed-effects: year: 11, source: 6
## Standard-errors: Clustered (author_name)
## Estimate Std. Error t value Pr(>|t|)
## elite_undergrad_ivyplusTRUE -0.024561 0.026565 -0.924568 0.35554
## edu.has_postgradTRUE -0.018263 0.020376 -0.896296 0.37043
## field.journoTRUE 0.034185 0.028846 1.185056 0.23643
## age_est_2017 > 30TRUE -0.048022 0.029278 -1.640241 0.10145
## gendermale -0.006272 0.020454 -0.306661 0.75920
## race.nonwhiteTRUE 0.006778 0.028647 0.236613 0.81303
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
## RMSE: 0.475843 Adj. R2: 0.013472
## Within R2: 0.00354Source competition
author.sources <- sources %>%
group_by(author_name, org_id) %>%
summarize(
n = n(),
)
author.sources.t <- sources %>%
group_by(author_name) %>%
summarize(total.n = n(), ideo.mean.i = mean(cfscore, na.rm = T))
a.s <- author.sources %>%
left_join(author.sources.t) %>%
mutate(
s = n / total.n * 100
)
a.s## # A tibble: 32,366 × 6
## # Groups: author_name [1,843]
## author_name org_id n total.n ideo.mean.i s
## <chr> <dbl> <int> <int> <dbl> <dbl>
## 1 abcarian, robin 9187 1 2 0.484 50
## 2 abcarian, robin NA 1 2 0.484 50
## 3 abderholden, frank 3450 1 9 -0.0976 11.1
## 4 abderholden, frank 9231 1 9 -0.0976 11.1
## 5 abderholden, frank 10050 4 9 -0.0976 44.4
## 6 abderholden, frank 14376 1 9 -0.0976 11.1
## 7 abderholden, frank 16242 1 9 -0.0976 11.1
## 8 abderholden, frank NA 1 9 -0.0976 11.1
## 9 abdi latif dahir 4557 1 2 NaN 50
## 10 abdi latif dahir NA 1 2 NaN 50
## # ℹ 32,356 more rowslibrary(hhi)
library(purrr)
compute_hhi <- function(df) {
author_name <- df$author_name
hhi <- sum(df$s^2)
data.frame(author_name, hhi)
}
df_hhi <- a.s %>%
group_split(author_name, .keep = TRUE) %>%
map(compute_hhi) %>%
bind_rows() %>%
distinct()
nrow(df_hhi)## [1] 1843df_hhi$hhi %>% hist()
df_hhi$hhi %>% summary()## Min. 1st Qu. Median Mean 3rd Qu. Max.
## 137.1 1072.7 1836.7 2368.9 3055.6 10000.0authors <- authors %>%
left_join(df_hhi) %>%
left_join(a.s)
etable(
# Ivy league, postgrads, but also younger Js have more diverse sources
c(lapply(c("hhi", "ideo.mean.i"), \(y) {
feols(make_fmla(y, j_covars, ""), data = authors)
}),
list(feols(make_fmla('cfscore', j_covars, ""), data = sources))
)
)## model 1 model 2
## Dependent Var.: hhi ideo.mean.i
##
## Constant 2,223.3*** (91.05) -0.4262*** (0.0560)
## elite_undergrad_ivyplusTRUE -389.1*** (85.04) 0.0472 (0.0519)
## edu.has_postgradTRUE -171.3* (67.55) 0.0328 (0.0412)
## field.journoTRUE 55.22 (69.98) 0.0659 (0.0434)
## age_est_2017>30TRUE -188.5** (66.16) 0.1430*** (0.0402)
## gendermale 89.19 (63.76) 0.1759*** (0.0396)
## race.nonwhiteTRUE -141.3. (81.11) -0.0050 (0.0507)
## ___________________________ __________________ ___________________
## S.E. type IID IID
## Observations 1,877 1,070
## R2 0.02386 0.03657
## Adj. R2 0.02073 0.03113
##
## model 3
## Dependent Var.: cfscore
##
## Constant -0.2554*** (0.0215)
## elite_undergrad_ivyplusTRUE 0.0701*** (0.0152)
## edu.has_postgradTRUE -0.0022 (0.0127)
## field.journoTRUE 0.0347* (0.0158)
## age_est_2017>30TRUE 0.1305*** (0.0143)
## gendermale -0.0448*** (0.0126)
## race.nonwhiteTRUE -0.0543** (0.0206)
## ___________________________ ___________________
## S.E. type IID
## Observations 18,459
## R2 0.00687
## Adj. R2 0.00655
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1DiD (Control - younger, less educated journalists)
The idea behind this DiD specification is that Trump’s presidency “activated” latent slant. We test whether journalists who are demographieally more likely to believe in the “participant” press altered their coverage after Trump’s presidency compared to their counterparts.
Parallel trends holds, suggesting that these two groups of journalists did not behave differently before Trump was elected. But after, the younger and less educated journalists slanted their coverage leftward more than the older and more educated ones did.
make_did <- function(df, refyear = 2016) {
iq.model <- feols(
ideo.mean.i ~ field.journo + age_est_2017 + elite_undergrad_ivyplus + gender + race.nonwhite,
data = df
)
df$pred_ideo <- predict(iq.model, newdata = df)
df$treat <- df$pred_ideo < quantile(df$pred_ideo, na.rm = T, 0.5)
#
summary(iq.model)
df %>% ggplot(aes(x = age_est_2017, fill = treat)) +
geom_histogram()
df %>% ggplot(aes(x = field.journo, fill = treat)) +
geom_bar(stat = "count")
df %>% ggplot(aes(x = elite_undergrad_ivyplus, fill = treat)) +
facet_wrap(~field.journo) +
geom_bar(stat = "count")
df %>% ggplot(aes(x = elite_undergrad_ivyplus, fill = treat)) +
facet_wrap(~race.nonwhite) +
geom_bar(stat = "count")
raw.plt <- df %>%
filter(!is.na(treat)) %>%
ggplot(aes(x = date, y = as.numeric(ideo.mean.i), color = treat, fill = treat)) +
geom_smooth(se = T, method = "loess") +
geom_vline(xintercept = as.Date("2016/01/01"), lty = "dashed") +
theme_bw() +
theme(
panel.grid.major = element_blank(),
panel.grid.minor = element_blank()
) +
labs(
x = "Year",
y = "Mean article slant",
color = "Predicted Type"
) +
guides(
fill = "none"
) +
scale_color_hue(label = c("Neutral", "Participant"))
raw.plt.b <- df %>%
filter(!is.na(treat)) %>%
ggplot(aes(x = date, y = as.numeric(balance_all), color = treat, fill = treat)) +
geom_smooth(se = T, method = "loess") +
geom_vline(xintercept = as.Date("2016/01/01"), lty = "dashed") +
theme_bw() +
theme(
panel.grid.major = element_blank(),
panel.grid.minor = element_blank()
) +
labs(
x = "Year",
y = "Article Balance",
color = "Predicted Type"
) +
guides(
fill = "none"
) +
scale_color_hue(label = c("Neutral", "Participant"))
raw.plt.c <- df %>%
filter(!is.na(treat)) %>%
ggplot(aes(x = date, y = bal.diff_lr_normalized, color = treat, fill = treat)) +
geom_smooth(se = T, method = "loess") +
geom_vline(xintercept = as.Date("2016/01/01"), lty = "dashed") +
theme_bw() +
theme(
panel.grid.major = element_blank(),
panel.grid.minor = element_blank()
) +
labs(
x = "Year",
y = "Article Balance (Diff LR Normalized)",
color = "Predicted Type"
) +
guides(
fill = "none"
) +
scale_color_hue(label = c("Neutral", "Participant"))
raw.plt.d <- df %>%
filter(!is.na(treat)) %>%
ggplot(aes(x = date, y = n, color = treat, fill = treat)) +
geom_smooth(se = T, method = "loess") +
geom_vline(xintercept = as.Date("2016/01/01"), lty = "dashed") +
theme_bw() +
theme(
panel.grid.major = element_blank(),
panel.grid.minor = element_blank()
) +
labs(
x = "Year",
y = "Number of Sources",
color = "Predicted Type"
) +
guides(
fill = "none"
) +
scale_color_hue(label = c("Neutral", "Participant"))
raw.plt.e <- df %>%
filter(!is.na(treat)) %>%
ggplot(aes(x = date, y = n_unique_source_category, color = treat, fill = treat)) +
geom_smooth(se = T, method = "loess") +
geom_vline(xintercept = as.Date("2016/01/01"), lty = "dashed") +
theme_bw() +
theme(
panel.grid.major = element_blank(),
panel.grid.minor = element_blank()
) +
labs(
x = "Year",
y = "Number of Source Types",
color = "Predicted Type"
) +
guides(
fill = "none"
) +
scale_color_hue(label = c("Neutral", "Participant"))
df <- df %>% mutate(
es.treat = case_when(
treat == 0 ~ 0,
treat == 1 & year < refyear ~ 0,
TRUE ~ 1
)
)
event_study_model <- feols(
ideo.mean.i ~ i(year, es.treat, ref = refyear) | year + author_name, df
)
event_study_model_balance <- feols(
as.numeric(balance_all) ~ i(year, es.treat, ref = refyear) | year + author_name, df
)
twfe <- feols(ideo.mean.i ~ (year > refyear) * treat | year + author_name, df)
list(
"raw_trends_plot" = raw.plt,
"article_balance_plot" = raw.plt.b,
"article_balance_plot2" = raw.plt.c,
"n_plot" = raw.plt.d,
"n_cat_plot" = raw.plt.e,
"event_study_model" = event_study_model,
"event_study_model_balance" = event_study_model_balance,
"twfe_model" = twfe
)
}
did <- make_did(articles, 2016)
did$raw_trends_plot
ggsave(here("paper/figures/did-raw-trends.png"), width = 6, height = 4)
ggarrange(plotlist = list(
did$n_plot,
did$n_cat_plot,
did$article_balance_plot,
did$article_balance_plot2
), common.legend = T)
library(ggfixest)
ggiplot(did$event_study_model) + labs(
title = "",
y = "Effect on Article Slant",
x = "Year"
)
# Event study plot
ggsave(here("paper/figures/did-event-study-slant.png"), width = 8, height = 5)
ggiplot(did$event_study_model_balance) + labs(
title = "",
y = "Effect on Article Balance",
x = "Year"
)
ggsave(here("paper/figures/did-event-study-balance.png"), width = 8, height = 5)
# TWFE
etable(did$twfe)## did$twfe
## Dependent Var.: ideo.mean.i
##
## year>2016TRUE x treatTRUE -0.0533 (0.0302)
## Fixed-Effects: ----------------
## year Yes
## author_name Yes
## _________________________ ________________
## S.E.: Clustered by: year
## Observations 8,703
## R2 0.23417
## Within R2 0.00027
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1DiD (Control - more conservative journalists)
We can also run a DiD where the control group is pre-2016 conservative journalists. It suggests that previously liberal journalists actually became more conservative (or that previously conservative journalists became more liberal) during Trump’s presidency.
I’m not exactly sure what this shows. The results are consistent with an alternative explanation that the most liberal journalists in the 2012-2016 period are just followers of presidential cues - the liberal cohort returns to liberalness after Biden comes into office.
authors_pre2017 <- articles %>%
filter(year < 2017) %>%
group_by(author_name) %>%
summarize(
ideo = mean(ideo.mean.i, na.rm = T)
) %>%
filter(!is.na(ideo))
authors_pre2017$ideo %>% summary()## Min. 1st Qu. Median Mean 3rd Qu. Max.
## -1.47000 -0.51534 -0.25272 -0.24424 -0.02143 1.19392articles.2 <- articles %>%
left_join(authors_pre2017) %>%
mutate(
ideo_median = case_when(
ideo > median(articles$ideo.mean.i, na.rm = T) ~ "Conservative",
ideo <= median(articles$ideo.mean.i, na.rm = T) ~ "Liberal",
TRUE ~ "New"
),
treat = case_when(
ideo_median == "Liberal" ~ TRUE,
ideo_median == "Conservative" ~ FALSE,
TRUE ~ NA
)
)
articles.2 %>%
filter(ideo_median != "New") %>%
ggplot(aes(x = date, y = ideo.mean.i, color = ideo_median)) +
geom_smooth(se = T, method = "loess") +
geom_vline(xintercept = as.Date("2016/01/01"), lty = "dashed")
# TWFE
feols(ideo.mean.i ~ post2017 * treat | year + author_name, data = articles.2)## OLS estimation, Dep. Var.: ideo.mean.i
## Observations: 6,478
## Fixed-effects: year: 11, author_name: 603
## Standard-errors: Clustered (year)
## Estimate Std. Error t value Pr(>|t|)
## post2017TRUE:treatTRUE 0.228178 0.04137 5.51558 0.00025614 ***
## ... 2 variables were removed because of collinearity (post2017TRUE and treatTRUE)
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
## RMSE: 0.431888 Adj. R2: 0.153967
## Within R2: 0.008122feols(balance_all ~ post2017 * treat | year + author_name, data = articles.2)## OLS estimation, Dep. Var.: balance_all
## Observations: 6,357
## Fixed-effects: year: 11, author_name: 575
## Standard-errors: Clustered (year)
## Estimate Std. Error t value Pr(>|t|)
## post2017TRUE:treatTRUE 0.151388 0.04368 3.46584 0.0060632 **
## ... 2 variables were removed because of collinearity (post2017TRUE and treatTRUE)
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
## RMSE: 0.444967 Adj. R2: 0.127735
## Within R2: 0.003368# Event study
feols(ideo.mean.i ~ i(year, treat, ref = 2016) | year + author, articles.2) %>% coefplot()
More did specs
make_did2 <- function(df) {
iq.model <- feols(
ideo.mean.i ~ field.journo + age_est_2017 + elite_undergrad_ivyplus + gender + race.nonwhite,
data = df
)
df$pred_ideo <- predict(iq.model, newdata = df)
df$treat1 <- df$pred_ideo < quantile(df$pred_ideo, na.rm = T, 0.2)
df$treat2 <- df$pred_ideo > quantile(df$pred_ideo, na.rm = T, 0.8)
df$treat <- ifelse(df$treat1, "Most Liberal", ifelse(df$treat2, "Most Conservative", "Control"))
raw.plt <- df %>%
filter(!is.na(treat)) %>%
ggplot(aes(x = date, y = as.numeric(ideo.mean.i), color = treat, fill = treat)) +
geom_smooth(se = T, method = "loess") +
geom_vline(xintercept = as.Date("2016/01/01"), lty = "dashed") +
theme_bw() +
theme(
panel.grid.major = element_blank(),
panel.grid.minor = element_blank()
) +
labs(
x = "Year",
y = "Mean article slant",
color = "Predicted Type"
) +
guides(
fill = "none"
) #+
# scale_color_hue(label = c("Neutral", "Participant"))
raw.plt.b <- df %>%
filter(!is.na(treat)) %>%
ggplot(aes(x = date, y = as.numeric(balance_all), color = treat, fill = treat)) +
geom_smooth(se = T, method = "loess") +
geom_vline(xintercept = as.Date("2016/01/01"), lty = "dashed") +
theme_bw() +
theme(
panel.grid.major = element_blank(),
panel.grid.minor = element_blank()
) +
labs(
x = "Year",
y = "Article Balance",
color = "Predicted Type"
) +
guides(
fill = "none"
)
# scale_color_hue(label = c("Neutral", "Participant"))
event_study_model <- feols(
ideo.mean.i ~ i(year, treat, ref = 2015) | year + author_name, df
)
event_study_model_balance <- feols(
as.numeric(balance_all) ~ i(year, treat, ref = 2016) | year + author_name, df
)
twfe <- feols(ideo.mean.i ~ (year > 2015) * treat | year + author_name, df)
list(
"raw_trends_plot" = raw.plt,
"article_balance_plot" = raw.plt.b,
"event_study_model" = event_study_model,
"event_study_model_balance" = event_study_model_balance,
"twfe_model" = twfe
)
}
did <- make_did2(articles)
did$raw_trends_plot
did$article_balance_plot
# Event study plot
did$event_study_model## OLS estimation, Dep. Var.: ideo.mean.i
## Observations: 8,703
## Fixed-effects: year: 11, author_name: 858
## Standard-errors: Clustered (year)
## Estimate Std. Error t value Pr(>|t|)
## year::2012:treat::Control -0.016307 0.051576 -0.316184 0.758364
## year::2012:treat::Most Conservative -0.143528 0.067946 -2.112374 0.060799
## year::2013:treat::Control 0.189849 0.096324 1.970953 0.077022
## year::2013:treat::Most Conservative -0.024784 0.080583 -0.307556 0.764730
## year::2014:treat::Control -0.096048 0.053158 -1.806843 0.100921
## year::2014:treat::Most Conservative -0.200043 0.064215 -3.115188 0.010963
## year::2016:treat::Control -0.728755 42766.307803 -0.000017 0.999987
## year::2016:treat::Most Conservative -0.684842 42766.287709 -0.000016 0.999988
##
## year::2012:treat::Control
## year::2012:treat::Most Conservative .
## year::2013:treat::Control .
## year::2013:treat::Most Conservative
## year::2014:treat::Control
## year::2014:treat::Most Conservative *
## year::2016:treat::Control
## year::2016:treat::Most Conservative
## ... 13 coefficients remaining (display them with summary() or use argument n)
## ... 9 variables were removed because of collinearity (year::2012:treat::Most Liberal, year::2013:treat::Most Liberal and 7 others [full set in $collin.var])
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
## RMSE: 0.423671 Adj. R2: 0.150296
## Within R2: 0.003974df <- articles
m.post <- feols(
ideo.mean.i ~ field.journo + age_est_2017 + elite_undergrad_ivyplus + gender + race.nonwhite | year + source,
data = df %>% filter(year > 2016)
)
m.pre <- feols(
ideo.mean.i ~ field.journo + age_est_2017 + elite_undergrad_ivyplus + gender + race.nonwhite | year + source,
data = df %>% filter(year <= 2016)
)
m.all <- feols(
ideo.mean.i ~ field.journo + age_est_2017 + elite_undergrad_ivyplus + gender + race.nonwhite | year + source,
data = df
)
m.all.nofe <- feols(
ideo.mean.i ~ field.journo + age_est_2017 + elite_undergrad_ivyplus + gender + race.nonwhite,
data = df
)
etable(list(m.pre, m.post, m.all, m.all.nofe))## model 1 model 2
## Dependent Var.: ideo.mean.i ideo.mean.i
##
## field.journoTRUE 0.0667 (0.0328) 0.0642** (0.0137)
## age_est_2017 -0.0012 (0.0016) 0.0018 (0.0012)
## elite_undergrad_ivyplusTRUE 0.0197 (0.0304) 0.1071* (0.0278)
## gendermale -0.0554 (0.0559) -0.0065 (0.0192)
## race.nonwhiteTRUE -0.0800. (0.0342) -0.0607* (0.0232)
## Constant
## Fixed-Effects: ----------------- -----------------
## year Yes Yes
## source Yes Yes
## ___________________________ _________________ _________________
## S.E. type by: year by: year
## Observations 1,740 6,963
## R2 0.05382 0.07361
## Within R2 0.01137 0.01499
##
## model 3 model 4
## Dependent Var.: ideo.mean.i ideo.mean.i
##
## field.journoTRUE 0.0657*** (0.0132) 0.0755*** (0.0127)
## age_est_2017 0.0014 (0.0010) 0.0015** (0.0005)
## elite_undergrad_ivyplusTRUE 0.0927** (0.0223) 0.0610*** (0.0123)
## gendermale -0.0167 (0.0219) -0.0306** (0.0110)
## race.nonwhiteTRUE -0.0663** (0.0198) -0.0789*** (0.0164)
## Constant -0.3174*** (0.0214)
## Fixed-Effects: ------------------ -------------------
## year Yes No
## source Yes No
## ___________________________ __________________ ___________________
## S.E. type by: year IID
## Observations 8,703 8,703
## R2 0.06780 0.01054
## Within R2 0.01230 --
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1# Could do articles about trump vs articles not about trump?Appendix
Coverage/data quality
How many articles do we have journalist education for?
articles.with.author <- articles %>% filter(!is.na(author))
articles.with.author.cfscore <- articles.with.author %>% filter(!is.na(ideo.mean.i))
nrow(articles.with.author)## [1] 12348nrow(articles.with.author.cfscore)## [1] 11314table(!is.na(articles.with.author$edu.undergrad)) %>% prop.table()##
## FALSE TRUE
## 0.1187237 0.8812763table(!is.na(articles.with.author$age_est)) %>% prop.table()##
## FALSE TRUE
## 0.08317136 0.91682864with(
articles.with.author.cfscore,
table(!is.na(age_est) & !is.na(elite_undergrad_ivyplus) & !is.na(edu.has_postgrad) & !is.na(field.journo) & !is.na(gender) & !is.na(race.nonwhite))
)##
## FALSE TRUE
## 2678 8636nrow(articles.with.author.cfscore %>% filter(is.na(gender)))## [1] 290nrow(articles.with.author.cfscore %>% filter(is.na(race.nonwhite)))## [1] 96nrow(articles.with.author.cfscore %>% filter(is.na(field.journo)))## [1] 2462nrow(articles.with.author.cfscore %>% filter(is.na(edu.has_postgrad)))## [1] 826nrow(articles.with.author.cfscore %>% filter(is.na(elite_undergrad_ivyplus)))## [1] 1297nrow(articles.with.author.cfscore %>% filter(is.na(age_est)))## [1] 908articles.with.author %>%
filter(
is.na(age_est) | is.na(elite_undergrad_ivyplus) | is.na(edu.has_postgrad) | is.na(field.journo) | is.na(gender) | is.na(race.nonwhite)
) %>%
group_by(
author_name, edu.undergrad, edu.field, edu.grad_year, edu.has_postgrad, exp.year_start, gender, race.nonwhite
) %>%
summarize(n = n(), newspapers = paste0(unique(source), collapse = ",")) %>%
arrange(desc(n)) %>%
write_tsv("manual_01-07-25.tsv")How many articles do we have journalist age for?
How many sources do we have DIME scores for?
Proportion of all sources:
table(sources$cfscore.impute %>% is.na, useNA = "ifany") %>% prop.table()##
## FALSE TRUE
## 0.5452656 0.4547344nrow(sources %>% filter(!is.na(cfscore))) / nrow(sources)## [1] 0.3082604table(
!is.na(sources$cfscore),
sources$category.slant
)##
## Academic Advocacy Bureaucrat Business Democrat Environmental
## FALSE 7506 2981 9635 3647 3658 7638
## TRUE 0 2384 0 2894 7549 4265
##
## Fossil Fuel International Media Other Politician Republican
## FALSE 1180 6829 4318 3955 1578 653
## TRUE 2422 46 0 695 0 3621Proportion of non government organizations:
n_eligible <- sources %>%
filter(tolower(category.slant) %in% c("advocacy", "environmental", "fossil fuel", "business", "democrat", "republican")) %>%
nrow()
# Excludes bureaucrat, media, other
nrow(sources %>% filter(!is.na(cfscore))) / n_eligible## [1] 0.5566539Histogram of measured article ideology
articles %>% ggplot(aes(x = ideo.mean.i, color = source, fill = source)) +
geom_density(alpha = 0.1) +
geom_hline(yintercept = 0) +
theme_bw() +
theme(panel.grid = element_blank()) +
labs(x = "Ideology", y = "Density")
ggsave(here("paper/figures/article-ideology-hist.png"), width = 6, height = 4)Histogram of predicted journalist ideology
articles %>%
filter(!is.na(pred.ideo.q)) %>%
group_by(author_name) %>%
summarize(pred.ideo = first(pred.ideo), pred.ideo.q = first(pred.ideo.q)) %>%
ggplot(aes(x = pred.ideo, fill = pred.ideo.q)) +
geom_histogram() +
geom_hline(yintercept = 0) +
theme_bw() +
theme(panel.grid = element_blank()) +
labs(x = "Predicted Ideology", y = "Count", fill = "Quartile")
ggsave(here("paper/figures/journo-pred-ideology-hist.png"), width = 6, height = 4)
articles %>% ggplot(aes(x = pred.ideo, y = ideo.mean.i, color = pred.ideo.q)) +
geom_point()
Distribution of source types by newspaper
sources %>%
mutate(
category = factor(category, levels = c(
"Academic",
"Advocacy",
"Business",
"Bureaucrat",
"Politician",
"International",
"Media",
"Other"
))
) %>%
group_by(category, source) %>%
summarize(n = n()) %>%
group_by(source) %>%
mutate(prop = n / sum(n)) %>%
ungroup() %>%
ggplot(
aes(x = reorder(source, n), fill = fct_rev(category), y = prop)
) +
geom_bar(stat = "identity") +
theme_bw() +
theme(panel.grid.major = element_blank(), panel.grid.minor = element_blank(), legend.position = "bottom") +
labs(x = "", y = "Prop", fill = "Category") +
scale_fill_brewer(palette = "Set2", direction = -1) +
coord_flip() +
guides(fill = guide_legend(reverse = TRUE))
ggsave(here("paper/figures/source-dist.png"), width = 8, height = 6)
sources.s2 <- sources %>%
mutate(category.slant2 = case_when(
category.slant %in% c("Bureaucrat", "International", "Politician") ~ "Other",
category.slant %in% c("Academic", "Media", "Advocacy") ~ "Other",
TRUE ~ category.slant
)) %>%
mutate(
category.slant2 = factor(category.slant2, levels = c(
"Environmental",
"Business",
"Fossil Fuel",
"Democrat",
"Republican",
"Other"
))
)
sources.s2 %>%
group_by(category.slant2, source) %>%
summarize(n = n()) %>%
group_by(source) %>%
mutate(prop = n / sum(n)) %>%
ungroup() %>%
ggplot(
aes(x = reorder(source, n), fill = fct_rev(category.slant2), y = prop)
) +
geom_bar(stat = "identity") +
scale_fill_manual(values = c(
"Environmental" = "#91bfdb",
"Democrat" = "#4575b4",
"Republican" = "#d73027",
"Fossil Fuel" = "#fc8d59",
"Business" = "#fee090",
"Other" = "#ffffbf"
)) +
theme_bw() +
theme(panel.grid.major = element_blank(), panel.grid.minor = element_blank(), legend.position = "bottom") +
labs(x = "", y = "Prop", fill = "Category") +
coord_flip() +
guides(fill = guide_legend(reverse = TRUE))
ggsave(here("paper/figures/source-dist-2.png"), width = 8, height = 6)
sources.s2$category.slant2 %>%
table() %>%
prop.table()## .
## Environmental Business Fossil Fuel Democrat Republican
## 0.15367831 0.08445013 0.04650502 0.14469233 0.05518114
## Other
## 0.51549307sources.s2 %>%
group_by(category.slant2) %>%
summarize(cfscore = mean(cfscore, na.rm = T))## # A tibble: 6 × 2
## category.slant2 cfscore
## <fct> <dbl>
## 1 Environmental -0.821
## 2 Business 0.105
## 3 Fossil Fuel 0.347
## 4 Democrat -0.573
## 5 Republican 0.895
## 6 Other -0.0914Bivariate analysis
models <- list(
feols(ideo.mean.i ~ elite_undergrad_ivyplus, data = articles),
feols(ideo.mean.i ~ edu.has_postgrad, data = articles),
feols(ideo.mean.i ~ field.journo, data = articles),
feols(ideo.mean.i ~ age_est > 40, data = articles),
feols(ideo.mean.i ~ gender, data = articles),
feols(ideo.mean.i ~ !race.nonwhite, data = articles),
feols(n_ff / (n_ff + n_env) ~ elite_undergrad_ivyplus, data = articles),
feols(n_ff / (n_ff + n_env) ~ edu.has_postgrad, data = articles),
feols(n_ff / (n_ff + n_env) ~ field.journo, data = articles),
feols(n_ff / (n_ff + n_env) ~ age_est > 40, data = articles),
feols(n_ff / (n_ff + n_env) ~ gender, data = articles),
feols(n_ff / (n_ff + n_env) ~ !race.nonwhite, data = articles),
feols(n_rep / (n_rep + n_dem) ~ elite_undergrad_ivyplus, data = articles),
feols(n_rep / (n_rep + n_dem) ~ edu.has_postgrad, data = articles),
feols(n_rep / (n_rep + n_dem) ~ field.journo, data = articles),
feols(n_rep / (n_rep + n_dem) ~ age_est > 40, data = articles),
feols(n_rep / (n_rep + n_dem) ~ gender, data = articles),
feols(n_rep / (n_rep + n_dem) ~ !race.nonwhite, data = articles),
feols(ideo.mean ~ elite_undergrad_ivyplus, data = articles),
feols(ideo.mean ~ edu.has_postgrad, data = articles),
feols(ideo.mean ~ field.journo, data = articles),
feols(ideo.mean ~ age_est > 40, data = articles),
feols(ideo.mean ~ gender, data = articles),
feols(ideo.mean ~ !race.nonwhite, data = articles)
)
res <- lapply(models, \(m) {
beta <- m$coeftable[2, 1]
sig <- m$coeftable[2, 4] < 0.05
xname <- rownames(m$coeftable)[[2]]
yname <- as.character(m$fml)[2]
list(
"beta" = beta,
"sig" = sig,
"x" = xname,
"y" = yname
)
}) %>% bind_rows()
res %>%
mutate(
beta = round(beta, 2),
fc = case_when(
sig & beta > 0 ~ "coral",
sig & beta < 0 ~ "cyan",
TRUE ~ "grey"
)
) %>%
ggplot(aes(x = y, y = x, fill = fc, label = beta)) +
geom_tile() +
geom_text(size = 8) +
scale_fill_manual(values = c("coral", "cyan3", "grey"))
Non policy articles cite fewer ideological sources
all.sources$cfscore_std <- all.sources$cfscore - mean(all.sources$cfscore, na.rm = T)
models <-
list(
lm(n_env + n_ff ~ policy_label_gpt, data = all.articles),
lm(n_rep + n_dem ~ policy_label_gpt, data = all.articles)
)
stargazer(models, type = "html")| Dependent variable: | ||
| n_env + n_ff | n_rep + n_dem | |
| (1) | (2) | |
| policy_label_gpt | 0.491*** | 0.825*** |
| (0.015) | (0.016) | |
| Constant | 0.689*** | 0.416*** |
| (0.010) | (0.011) | |
| Observations | 28,121 | 28,121 |
| R2 | 0.036 | 0.083 |
| Adjusted R2 | 0.036 | 0.083 |
| Residual Std. Error (df = 28119) | 1.261 | 1.363 |
| F Statistic (df = 1; 28119) | 1,063.074*** | 2,561.112*** |
| Note: | p<0.1; p<0.05; p<0.01 | |
stargazer(models)
% Table created by stargazer v.5.2.3 by Marek Hlavac, Social Policy Institute. E-mail: marek.hlavac at gmail.com
% Date and time: Sat, Feb 08, 2025 - 16:19:53
\begin{table}[!htbp] \centering
\caption{}
\label{}
\begin{tabular}{@{\extracolsep{5pt}}lcc}
\\[-1.8ex]\hline
\hline \\[-1.8ex]
& \multicolumn{2}{c}{\textit{Dependent variable:}} \\
\cline{2-3}
\\[-1.8ex] & n\_env + n\_ff & n\_rep + n\_dem \\
\\[-1.8ex] & (1) & (2)\\
\hline \\[-1.8ex]
policy\_label\_gpt & 0.491$^{***}$ & 0.825$^{***}$ \\
& (0.015) & (0.016) \\
& & \\
Constant & 0.689$^{***}$ & 0.416$^{***}$ \\
& (0.010) & (0.011) \\
& & \\
\hline \\[-1.8ex]
Observations & 28,121 & 28,121 \\
R$^{2}$ & 0.036 & 0.083 \\
Adjusted R$^{2}$ & 0.036 & 0.083 \\
Residual Std. Error (df = 28119) & 1.261 & 1.363 \\
F Statistic (df = 1; 28119) & 1,063.074$^{***}$ & 2,561.112$^{***}$ \\
\hline
\hline \\[-1.8ex]
\textit{Note:} & \multicolumn{2}{r}{$^{*}$p$<$0.1; $^{**}$p$<$0.05; $^{***}$p$<$0.01} \\
\end{tabular}
\end{table} Alternate Specifications
models <- make_models(all.articles, c("ideo.mean.i", "balance_all"), j_covars)
models2 <- make_models(articles, c("ideo.mean", "-prop_topic_environment"), j_covars)
etable(models2)## model 1 model 2
## Dependent Var.: ideo.mean -prop_topic_environment
##
## elite_undergrad_ivyplusTRUE 0.0970*** (0.0290) 0.0813*** (0.0154)
## edu.has_postgradTRUE -0.0307 (0.0248) -0.0162 (0.0127)
## field.journoTRUE 0.0340 (0.0328) 0.0664*** (0.0174)
## age_est_2017>30TRUE 0.0555* (0.0250) 0.0129 (0.0189)
## gendermale -0.0083 (0.0231) -0.0005 (0.0132)
## race.nonwhiteTRUE -0.0637. (0.0384) -0.0634** (0.0210)
## Fixed-Effects: ------------------ -----------------------
## year Yes Yes
## source Yes Yes
## ___________________________ __________________ _______________________
## S.E.: Clustered by: author_name by: author_name
## Observations 7,051 9,349
## R2 0.05908 0.08660
## Within R2 0.00800 0.04112
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1etable(c(models, models2))## model 1 model 2
## Dependent Var.: ideo.mean.i balance_all
##
## elite_undergrad_ivyplusTRUE 0.0526** (0.0165) 0.1722*** (0.0287)
## edu.has_postgradTRUE -0.0101 (0.0159) -0.0326 (0.0275)
## field.journoTRUE 0.0444* (0.0183) 0.0998*** (0.0251)
## age_est_2017>30TRUE 0.0417** (0.0146) 0.0503. (0.0275)
## gendermale -0.0032 (0.0148) -0.0019 (0.0255)
## race.nonwhiteTRUE -0.0501* (0.0223) -0.0666. (0.0340)
## Fixed-Effects: ----------------- ------------------
## year Yes Yes
## source Yes Yes
## ___________________________ _________________ __________________
## S.E.: Clustered by: author_name by: author_name
## Observations 15,008 13,699
## R2 0.04795 0.05041
## Within R2 0.00654 0.03303
##
## model 3 model 4
## Dependent Var.: ideo.mean -prop_topic_environment
##
## elite_undergrad_ivyplusTRUE 0.0970*** (0.0290) 0.0813*** (0.0154)
## edu.has_postgradTRUE -0.0307 (0.0248) -0.0162 (0.0127)
## field.journoTRUE 0.0340 (0.0328) 0.0664*** (0.0174)
## age_est_2017>30TRUE 0.0555* (0.0250) 0.0129 (0.0189)
## gendermale -0.0083 (0.0231) -0.0005 (0.0132)
## race.nonwhiteTRUE -0.0637. (0.0384) -0.0634** (0.0210)
## Fixed-Effects: ------------------ -----------------------
## year Yes Yes
## source Yes Yes
## ___________________________ __________________ _______________________
## S.E.: Clustered by: author_name by: author_name
## Observations 7,051 9,349
## R2 0.05908 0.08660
## Within R2 0.00800 0.04112
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1did <- make_did(all.articles)
pdf(here("paper/figures/did-event-study-alt.pdf"))
did$event_study_model %>%
coefplot(main = "", xlab = "")
dev.off()## quartz_off_screen
## 2etable(c(models, models2), tex = T, digits = 2)\begingroup
\centering
\begin{tabular}{lcccc}
\tabularnewline \midrule \midrule
Dependent Variables: & ideo.mean.i & balance\_all & ideo.mean & -prop\_topic\_environment\\
Model: & (1) & (2) & (3) & (4)\\
\midrule
\emph{Variables}\\
elite\_undergrad\_ivyplusTRUE & 0.05$^{***}$ & 0.17$^{***}$ & 0.10$^{***}$ & 0.08$^{***}$\\
& (0.02) & (0.03) & (0.03) & (0.01)\\
edu.has\_postgradTRUE & -0.01 & -0.03 & -0.03 & -0.02\\
& (0.02) & (0.03) & (0.03) & (0.01)\\
field.journoTRUE & 0.04$^{**}$ & 0.10$^{***}$ & 0.03 & 0.07$^{***}$\\
& (0.02) & (0.03) & (0.03) & (0.02)\\
age\_est\_2017>30TRUE & 0.04$^{***}$ & 0.05$^{*}$ & 0.06$^{**}$ & 0.01\\
& (0.01) & (0.03) & (0.03) & (0.02)\\
gendermale & -0.003 & -0.002 & -0.008 & -0.0005\\
& (0.01) & (0.03) & (0.02) & (0.01)\\
race.nonwhiteTRUE & -0.05$^{**}$ & -0.07$^{*}$ & -0.06$^{*}$ & -0.06$^{***}$\\
& (0.02) & (0.03) & (0.04) & (0.02)\\
\midrule
\emph{Fixed-effects}\\
year & Yes & Yes & Yes & Yes\\
source & Yes & Yes & Yes & Yes\\
\midrule
\emph{Fit statistics}\\
Observations & 15,008 & 13,699 & 7,051 & 9,349\\
R$^2$ & 0.04795 & 0.05041 & 0.05908 & 0.08660\\
Within R$^2$ & 0.00654 & 0.03303 & 0.00800 & 0.04112\\
\midrule \midrule
\multicolumn{5}{l}{\emph{Clustered (author\_name) standard-errors in parentheses}}\\
\multicolumn{5}{l}{\emph{Signif. Codes: ***: 0.01, **: 0.05, *: 0.1}}\\
\end{tabular}
\par\endgroupInvestigating Drivers
models3 <- make_models(articles, c(
"ideo.mean.pols",
"ideo.mean.i.orgs",
"n_ff",
"n_env",
"n_rep",
"n_dem",
"n_right_dime",
"-n_left_dime",
"(n_ff > 0)",
"(n_env > 0)",
"(n_rep > 0)",
"(n_dem > 0)",
"prop_ff",
"prop_env",
"prop_rep",
"prop_dem"
), j_covars)
etable(models3)## model 1 model 2
## Dependent Var.: ideo.mean.pols ideo.mean.i.orgs
##
## elite_undergrad_ivyplusTRUE 0.0507 (0.0378) 0.1117** (0.0403)
## edu.has_postgradTRUE -0.0467 (0.0321) -0.0325 (0.0373)
## field.journoTRUE -0.0035 (0.0429) -0.0030 (0.0469)
## age_est_2017>30TRUE 0.1140*** (0.0296) 0.0346 (0.0394)
## gendermale -0.0379 (0.0301) 0.0194 (0.0335)
## race.nonwhiteTRUE -0.0992* (0.0473) -0.0416 (0.0684)
## Fixed-Effects: ------------------ -----------------
## year Yes Yes
## source Yes Yes
## ___________________________ __________________ _________________
## S.E.: Clustered by: author_name by: author_name
## Observations 4,492 4,956
## R2 0.09378 0.05159
## Within R2 0.00989 0.00733
##
## model 3 model 4
## Dependent Var.: n_ff n_env
##
## elite_undergrad_ivyplusTRUE 0.0953 (0.0650) 0.0776 (0.0775)
## edu.has_postgradTRUE -0.0184 (0.0616) 0.0741 (0.0607)
## field.journoTRUE 0.0123 (0.0726) -0.1549* (0.0646)
## age_est_2017>30TRUE 0.0274 (0.0580) -0.0116 (0.0876)
## gendermale 0.0124 (0.0561) 0.0489 (0.0626)
## race.nonwhiteTRUE -0.0347 (0.0760) 0.2569** (0.0828)
## Fixed-Effects: ---------------- -----------------
## year Yes Yes
## source Yes Yes
## ___________________________ ________________ _________________
## S.E.: Clustered by: author_name by: author_name
## Observations 9,349 9,349
## R2 0.02671 0.04064
## Within R2 0.00396 0.01189
##
## model 5 model 6
## Dependent Var.: n_rep n_dem
##
## elite_undergrad_ivyplusTRUE 0.1879*** (0.0534) 0.3820*** (0.1132)
## edu.has_postgradTRUE -0.0834 (0.0526) -0.1297 (0.1144)
## field.journoTRUE 0.1483** (0.0461) 0.4698*** (0.1127)
## age_est_2017>30TRUE 0.0516 (0.0568) -0.1656 (0.1689)
## gendermale -0.0727 (0.0451) -0.0575 (0.0980)
## race.nonwhiteTRUE -0.0998* (0.0502) -0.1762. (0.1039)
## Fixed-Effects: ------------------ ------------------
## year Yes Yes
## source Yes Yes
## ___________________________ __________________ __________________
## S.E.: Clustered by: author_name by: author_name
## Observations 9,349 9,349
## R2 0.06115 0.09073
## Within R2 0.02433 0.04155
##
## model 7 model 8
## Dependent Var.: n_right_dime -n_left_dime
##
## elite_undergrad_ivyplusTRUE 0.2522*** (0.0552) -0.2356** (0.0857)
## edu.has_postgradTRUE -0.1339* (0.0540) 0.0943 (0.0823)
## field.journoTRUE 0.1110* (0.0534) -0.2506** (0.0917)
## age_est_2017>30TRUE 0.0394 (0.0582) 0.1425 (0.1169)
## gendermale -0.0340 (0.0476) -0.0304 (0.0822)
## race.nonwhiteTRUE -0.1279. (0.0716) 0.0582 (0.0769)
## Fixed-Effects: ------------------ ------------------
## year Yes Yes
## source Yes Yes
## ___________________________ __________________ __________________
## S.E.: Clustered by: author_name by: author_name
## Observations 9,349 9,349
## R2 0.04070 0.06299
## Within R2 0.02309 0.01600
##
## model 9 model 10
## Dependent Var.: n_ff>0 n_env>0
##
## elite_undergrad_ivyplusTRUE 0.0677* (0.0307) 0.0416 (0.0270)
## edu.has_postgradTRUE 0.0007 (0.0282) 0.0199 (0.0238)
## field.journoTRUE 0.0296 (0.0330) -0.0787** (0.0277)
## age_est_2017>30TRUE 0.0187 (0.0284) 0.0265 (0.0328)
## gendermale -0.0082 (0.0269) 0.0033 (0.0236)
## race.nonwhiteTRUE -0.0283 (0.0343) 0.0911** (0.0316)
## Fixed-Effects: ---------------- ------------------
## year Yes Yes
## source Yes Yes
## ___________________________ ________________ __________________
## S.E.: Clustered by: author_name by: author_name
## Observations 9,349 9,349
## R2 0.03022 0.04237
## Within R2 0.00717 0.01269
##
## model 11 model 12
## Dependent Var.: n_rep>0 n_dem>0
##
## elite_undergrad_ivyplusTRUE 0.1159*** (0.0322) 0.1925*** (0.0418)
## edu.has_postgradTRUE -0.0504 (0.0319) -0.0053 (0.0332)
## field.journoTRUE 0.0850** (0.0273) 0.1736*** (0.0339)
## age_est_2017>30TRUE 0.0328 (0.0327) 0.0042 (0.0408)
## gendermale -0.0494. (0.0281) -0.0283 (0.0333)
## race.nonwhiteTRUE -0.0617* (0.0304) -0.0377 (0.0398)
## Fixed-Effects: ------------------ ------------------
## year Yes Yes
## source Yes Yes
## ___________________________ __________________ __________________
## S.E.: Clustered by: author_name by: author_name
## Observations 9,349 9,349
## R2 0.06748 0.07851
## Within R2 0.02975 0.03916
##
## model 13 model 14
## Dependent Var.: prop_ff prop_env
##
## elite_undergrad_ivyplusTRUE 0.0134 (0.0097) 0.0019 (0.0100)
## edu.has_postgradTRUE -0.0008 (0.0096) 0.0107 (0.0091)
## field.journoTRUE 0.0048 (0.0126) -0.0243* (0.0105)
## age_est_2017>30TRUE 0.0078 (0.0081) -0.0022 (0.0120)
## gendermale 0.0010 (0.0088) 0.0079 (0.0089)
## race.nonwhiteTRUE -0.0070 (0.0108) 0.0452** (0.0143)
## Fixed-Effects: ---------------- -----------------
## year Yes Yes
## source Yes Yes
## ___________________________ ________________ _________________
## S.E.: Clustered by: author_name by: author_name
## Observations 9,349 9,349
## R2 0.03902 0.03227
## Within R2 0.00338 0.01113
##
## model 15 model 16
## Dependent Var.: prop_rep prop_dem
##
## elite_undergrad_ivyplusTRUE 0.0280** (0.0085) 0.0541* (0.0215)
## edu.has_postgradTRUE -0.0130 (0.0086) -0.0081 (0.0196)
## field.journoTRUE 0.0281*** (0.0079) 0.0880*** (0.0210)
## age_est_2017>30TRUE 0.0080 (0.0085) -0.0128 (0.0274)
## gendermale -0.0127. (0.0074) -0.0123 (0.0169)
## race.nonwhiteTRUE -0.0171* (0.0082) -0.0305 (0.0205)
## Fixed-Effects: ------------------ ------------------
## year Yes Yes
## source Yes Yes
## ___________________________ __________________ __________________
## S.E.: Clustered by: author_name by: author_name
## Observations 9,349 9,349
## R2 0.06281 0.07175
## Within R2 0.02249 0.03117
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1etable(models3, tex = T)\begingroup
\centering
\begin{tabular}{lcccccccccccccccc}
\tabularnewline \midrule \midrule
Dependent Variables: & ideo.mean.pols & ideo.mean.i.orgs & n\_ff & n\_env & n\_rep & n\_dem & n\_right\_dime & -n\_left\_dime & n\_ff>0 & n\_env>0 & n\_rep>0 & n\_dem>0 & prop\_ff & prop\_env & prop\_rep & prop\_dem\\
Model: & (1) & (2) & (3) & (4) & (5) & (6) & (7) & (8) & (9) & (10) & (11) & (12) & (13) & (14) & (15) & (16)\\
\midrule
\emph{Variables}\\
elite\_undergrad\_ivyplusTRUE & 0.0507 & 0.1117$^{***}$ & 0.0953 & 0.0776 & 0.1879$^{***}$ & 0.3820$^{***}$ & 0.2522$^{***}$ & -0.2356$^{***}$ & 0.0677$^{**}$ & 0.0416 & 0.1159$^{***}$ & 0.1925$^{***}$ & 0.0134 & 0.0019 & 0.0280$^{***}$ & 0.0541$^{**}$\\
& (0.0378) & (0.0403) & (0.0650) & (0.0775) & (0.0534) & (0.1132) & (0.0552) & (0.0857) & (0.0307) & (0.0270) & (0.0322) & (0.0418) & (0.0097) & (0.0100) & (0.0085) & (0.0215)\\
edu.has\_postgradTRUE & -0.0467 & -0.0325 & -0.0184 & 0.0741 & -0.0834 & -0.1297 & -0.1339$^{**}$ & 0.0943 & 0.0007 & 0.0199 & -0.0504 & -0.0053 & -0.0008 & 0.0107 & -0.0130 & -0.0081\\
& (0.0321) & (0.0373) & (0.0616) & (0.0607) & (0.0526) & (0.1144) & (0.0540) & (0.0823) & (0.0282) & (0.0238) & (0.0319) & (0.0332) & (0.0096) & (0.0091) & (0.0086) & (0.0196)\\
field.journoTRUE & -0.0035 & -0.0030 & 0.0123 & -0.1549$^{**}$ & 0.1483$^{***}$ & 0.4698$^{***}$ & 0.1110$^{**}$ & -0.2506$^{***}$ & 0.0296 & -0.0787$^{***}$ & 0.0850$^{***}$ & 0.1736$^{***}$ & 0.0048 & -0.0243$^{**}$ & 0.0281$^{***}$ & 0.0880$^{***}$\\
& (0.0429) & (0.0469) & (0.0726) & (0.0646) & (0.0461) & (0.1127) & (0.0534) & (0.0917) & (0.0330) & (0.0277) & (0.0273) & (0.0339) & (0.0126) & (0.0105) & (0.0079) & (0.0210)\\
age\_est\_2017>30TRUE & 0.1140$^{***}$ & 0.0346 & 0.0274 & -0.0116 & 0.0516 & -0.1656 & 0.0394 & 0.1425 & 0.0187 & 0.0265 & 0.0328 & 0.0042 & 0.0078 & -0.0022 & 0.0080 & -0.0128\\
& (0.0296) & (0.0394) & (0.0580) & (0.0876) & (0.0568) & (0.1689) & (0.0582) & (0.1169) & (0.0284) & (0.0328) & (0.0327) & (0.0408) & (0.0081) & (0.0120) & (0.0085) & (0.0274)\\
gendermale & -0.0379 & 0.0194 & 0.0124 & 0.0489 & -0.0727 & -0.0575 & -0.0340 & -0.0304 & -0.0082 & 0.0033 & -0.0494$^{*}$ & -0.0283 & 0.0010 & 0.0079 & -0.0127$^{*}$ & -0.0123\\
& (0.0301) & (0.0335) & (0.0561) & (0.0626) & (0.0451) & (0.0980) & (0.0476) & (0.0822) & (0.0269) & (0.0236) & (0.0281) & (0.0333) & (0.0088) & (0.0089) & (0.0074) & (0.0169)\\
race.nonwhiteTRUE & -0.0992$^{**}$ & -0.0416 & -0.0347 & 0.2569$^{***}$ & -0.0998$^{**}$ & -0.1762$^{*}$ & -0.1279$^{*}$ & 0.0582 & -0.0283 & 0.0911$^{***}$ & -0.0617$^{**}$ & -0.0377 & -0.0070 & 0.0452$^{***}$ & -0.0171$^{**}$ & -0.0305\\
& (0.0473) & (0.0684) & (0.0760) & (0.0828) & (0.0502) & (0.1039) & (0.0716) & (0.0769) & (0.0343) & (0.0316) & (0.0304) & (0.0398) & (0.0108) & (0.0143) & (0.0082) & (0.0205)\\
\midrule
\emph{Fixed-effects}\\
year & Yes & Yes & Yes & Yes & Yes & Yes & Yes & Yes & Yes & Yes & Yes & Yes & Yes & Yes & Yes & Yes\\
source & Yes & Yes & Yes & Yes & Yes & Yes & Yes & Yes & Yes & Yes & Yes & Yes & Yes & Yes & Yes & Yes\\
\midrule
\emph{Fit statistics}\\
Observations & 4,492 & 4,956 & 9,349 & 9,349 & 9,349 & 9,349 & 9,349 & 9,349 & 9,349 & 9,349 & 9,349 & 9,349 & 9,349 & 9,349 & 9,349 & 9,349\\
R$^2$ & 0.09378 & 0.05159 & 0.02671 & 0.04064 & 0.06115 & 0.09073 & 0.04070 & 0.06299 & 0.03022 & 0.04237 & 0.06748 & 0.07851 & 0.03902 & 0.03227 & 0.06281 & 0.07175\\
Within R$^2$ & 0.00989 & 0.00733 & 0.00396 & 0.01189 & 0.02433 & 0.04155 & 0.02309 & 0.01600 & 0.00717 & 0.01269 & 0.02975 & 0.03916 & 0.00338 & 0.01113 & 0.02249 & 0.03117\\
\midrule \midrule
\multicolumn{17}{l}{\emph{Clustered (author\_name) standard-errors in parentheses}}\\
\multicolumn{17}{l}{\emph{Signif. Codes: ***: 0.01, **: 0.05, *: 0.1}}\\
\end{tabular}
\par\endgroupElite Educated Slant Over Time
articles %>%
ggplot(aes(x = date, y = ideo.mean.i, color = elite_undergrad_ivyplus, group = elite_undergrad_ivyplus)) +
geom_smooth(se = F, method = "loess") +
theme_bw() +
theme(panel.grid.minor = element_blank()) +
labs(title = "Ideology of sources over time by elite education")
NYT comment counts
nyt.articles <- articles %>% filter(source == "New York Times")
nyt.articles %>% ggplot(aes(x = date, y = n_comments, color = ideo.mean.i)) +
geom_point(alpha = 0.3, size = 10) +
scale_color_gradient(low = "blue", high = "red")
nyt.articles %>% ggplot(aes(x = date, y = n_comments, color = balance_all)) +
geom_point()
hist(nyt.articles$n_comments)
sd(nyt.articles$n_comments, na.rm = T)## [1] 512.5869table(nyt.articles$n_comments %>% is.na())##
## FALSE TRUE
## 1363 3308nyt.articles %>% ggplot(aes(x = ideo.mean.i, y = n_comments)) +
geom_point() +
geom_smooth() +
theme_bw()
ggsave(here("paper/figures/nyt-comments-slant-scatter.png"), width = 6, height = 5)
summary(feols(n_comments ~ ideo.mean.i, data = nyt.articles))## OLS estimation, Dep. Var.: n_comments
## Observations: 1,301
## Standard-errors: IID
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) 388.4150 15.8482 24.50848 < 2.2e-16 ***
## ideo.mean.i 92.9292 31.2870 2.97022 0.0030305 **
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
## RMSE: 518.7 Adj. R2: 0.005981m <- feols(n_comments ~ elite_undergrad_ivyplus + age_est_2017 + race.nonwhite + gender + field.journo + edu.has_postgrad, data = nyt.articles)
etable(m, tex = T)## \begingroup
## \centering
## \begin{tabular}{lc}
## \tabularnewline \midrule \midrule
## Dependent Variable: & n\_comments\\
## Model: & (1)\\
## \midrule
## \emph{Variables}\\
## Constant & 391.9$^{***}$\\
## & (99.39)\\
## elite\_undergrad\_ivyplusTRUE & 81.80$^{*}$\\
## & (48.56)\\
## age\_est\_2017 & 0.9671\\
## & (2.344)\\
## race.nonwhiteTRUE & -55.58\\
## & (62.47)\\
## gendermale & -25.65\\
## & (45.79)\\
## field.journoTRUE & 174.2$^{***}$\\
## & (54.27)\\
## edu.has\_postgradTRUE & -144.1$^{***}$\\
## & (38.99)\\
## \midrule
## \emph{Fit statistics}\\
## Observations & 978\\
## R$^2$ & 0.03599\\
## Adjusted R$^2$ & 0.03003\\
## \midrule \midrule
## \multicolumn{2}{l}{\emph{IID standard-errors in parentheses}}\\
## \multicolumn{2}{l}{\emph{Signif. Codes: ***: 0.01, **: 0.05, *: 0.1}}\\
## \end{tabular}
## \par\endgroupWhat articles get more comments?
summary(feols(n_comments ~ n + n_unique_source_category | year, data = nyt.articles)) summary(feols(log(n_comments) ~ n + n_unique_source_category | year, data = nyt.articles))
Journalist characteristics = journalist background is sort of significant I guess?
summary(feols(log(n_comments) ~ elite_undergrad_ivyplus + age_est_2017 + race.nonwhite + gender + field.journo + edu.has_postgrad, data = nyt.articles))
More comments for more rightward leaning articles.. that’s basically it
Not sure exactly how to interpret this…
summary(feols(n_comments ~ balance_all, data = nyt.articles)) summary(feols(n_comments ~ ideo.mean.i + balance_all, data = nyt.articles)) summary(feols(log(n_comments) ~ ideo.mean.i + balance_all, data = nyt.articles))
Interesting…
summary(feols(n_comments ~ ideo.mean.i + balance_all + n + n_unique_source_category, data = nyt.articles)) summary(feols(log(n_comments) ~ ideo.mean.i + balance_all + n + n_unique_source_category, data = nyt.articles)) summary(feols(n_comments ~ ideo.mean.i + balance_all + post2017, data = nyt.articles))
no super intersting interaction with the post
summary(feols(n_comments ~ (ideo.mean.i > 0) * post2017, data = nyt.articles)) summary(feols(log(n_comments) ~ (ideo.mean.i > 0) * post2017, data = nyt.articles))
nyt.articles %>% ggplot(aes(x = date, y = n_comments, color = balance_all)) + geom_smooth(se = F, method = “lm”) + geom_point() nyt.articles %>% ggplot(aes(x = date, y = n_comments, color = ideo.mean.i > 0)) + geom_smooth(se = F, method = “lm”) + geom_point()
etable( list( feols(n_comments ~ elite_undergrad_ivyplus + age_est_2017 + race.nonwhite + gender + field.journo + edu.has_postgrad, data = nyt.articles), # feols(n_comments ~ elite_undergrad_ivyplus + age_est_2017 + race.nonwhite + gender + field.journo + edu.has_postgrad | year, data = nyt.articles), feols(n_comments ~ ideo.mean.i + balance_all + n + n_unique_source_category, data = nyt.articles) # feols(n_comments ~ ideo.mean.i + balance_all + n + n_unique_source_category | year, data = nyt.articles) ) ) ```